Hydrogen uptake prediction in porous carbon materials explained by decision tree machine learning Algorithms: From experimental data to interpretable predictions

Widespread adoption of hydrogen fuel is constrained by the cost and safety limits of high-pressure and cryogenic storage. Adsorption-based storage in Porous Carbon Materials (PCMs) is a promising alternative, yet its potential is unrealized due to the research time and cost of discovery. A Machine L...

Full description

Saved in:
Bibliographic Details
Published in:International journal of hydrogen energy Vol. 197; p. 152704
Main Authors: Sunkara, Hemanth, Bhat A S, Shravani, R, Namitha, Acharya, Sushmitha, Shekar, Selva Kumar, Sainath, Krishnamurthy, Siddiqui, Shabnam
Format: Journal Article
Language:English
Published: Elsevier Ltd 05.01.2026
Subjects:
Adsorption
Decision trees
Hydrogen
Machine learning
Porous carbon materials
SHAP
SHAP
Adsorption
Porous carbon materials
Decision trees
Hydrogen
Machine learning
ISSN:0360-3199
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Widespread adoption of hydrogen fuel is constrained by the cost and safety limits of high-pressure and cryogenic storage. Adsorption-based storage in Porous Carbon Materials (PCMs) is a promising alternative, yet its potential is unrealized due to the research time and cost of discovery. A Machine Learning (ML) approach was developed using five Decision Tree-based models on a 2,101datapoint PCM dataset to rapidly address the demands of this gap. CatBoost delivered the best performance (R2 = 0.9983, RMSE = 0.094, and MAE = 0.053), outperforming the Stacking Ensemble model (improving R2 by 0.1 %, RMSE by 13 %, and MAE by 15 %). Further, SHAP analysis confirmed pressure, temperature, SBET, and pore volumes as the key predictors, aligning with adsorption theory. This ML strategy serves as an efficient pre-screening tool for accelerating PCM discovery and reducing research cost and time for safe and cost-effective hydrogen storage with higher interpretability compared to previously developed tools. [Display omitted] •Five Decision Tree ML models were used to predict H2 uptake in PCMs.•Gradient Boosting models showed stronger regression and residual performance.•CatBoost gave the best results with R2 of 0.9983 and RMSE of 0.094.•SHAP analysis found SBET as the most influential morphological property.•A Stacking ensemble outperformed its individual constituent models.
ISSN:0360-3199
DOI:10.1016/j.ijhydene.2025.152704